Culinary Hub

Team No.6
Table of contents

01 02
Introduction Literature Review

03 04
Brief Overview Conclusion
 TAILORED
RECIPES AND
CULINARY
COMMUNITY

Guided By : Prof.Syamamol T

Team Members :

Alenta Joseph
Vishnu A.V
Juval James
 INTRODUCTION

Welcome to the culinary future! Our recipe recommendation app isn't just a
tool; it's a culinary companion, fusing technology and taste. Committed to your
satisfaction, our platform personalizes recipe suggestions through a dynamic
machine learning algorithm that adapts to your culinary preferences and
dietary needs. Going beyond traditional recommendations, our project tackles
the nutrition challenge, aiming to accurately quantify the calorie content of food
items. In an era of heightened health consciousness, we empower you to
make informed dietary choices. Embark on a transformative culinary journey
with our machine-learning-fueled app. It's more than recipes; it's a discovery of
tailored dishes crafted to your tastes. Join us in reshaping the cooking
experience, where technology meets flavor, and every meal becomes an
adventure.
 Generating Personalized Recipes from Historical User Preferences(cseweb.ucsd.edu)
Bodhisattwa Prasad Majumder∗ , Shuyang Li∗ , Jianmo Ni, Julian McAuley

● Introduction:
● Users often rely on online recipes, but existing approaches neglect users with incomplete knowledge of ingredients.
● The paper proposes personalized recipe generation based on user preferences extracted from historical interactions with recipes.
● The model leverages an encoder-decoder framework and attends to user profiles to generate personalized recipes.
● Approach:
● The model takes the recipe name, partial list of ingredients, and calorie level as input and generates recipe instructions.
● Encoder-decoder framework with attention mechanisms is used, and user preferences are incorporated through attention fusion layers.
● Two schemes for modeling user preferences are explored: recipe interactions and techniques seen in historical data.
● Related Work:
● The paper draws on large-scale transformer-based language models for creative text generation.
● Recipe generation is framed as a data-to-text natural language generation problem, related to applications in automated journalism and summarization.
● Recipe Dataset:
● A novel dataset of 180K+ recipes and 700K+ user reviews is collected from Food.com.
● The model is trained on recipes with diverse lengths, ingredients, and user profiles.
● Experiments and Results:
● The proposed personalized models outperform baseline models in BPE perplexity and achieve better user matching accuracy and mean reciprocal rank.
● Personalized models show improvements in recipe-level coherence, local coherence, and are preferred by human evaluators in a pairwise comparison.
 Recipe Recommendation by Ingredients Detection(www.ijraset.com)
Mrs. Silviya D’monte1 , Department of Computer Engineering, Universal College of Engineering, Vasai

● Key Features:
● Ingredients Detection:
○ YOLOv5 CNN for easy ingredient identification.(covolutional neural network).
● Recipe Recommendation:
○ Hybrid model suggests recipes based on user history and preferences.
● Multilingual Support:
○ Offers recipes in various languages.
○ Enhances accessibility and cultural experience.
● Voice Assistance (NLP):
○ NLP for voice-controlled instructions.
○ Users can listen to recipes without device interaction.
○ System:
● Modules: user registration, detection, recommendation, hybrid model.
● YOLOv5: 87% accuracy with a custom dataset.
● Firebase: Real-time database storage.
● User Interface:
● Intuitive design with swipe cards for easy browsing.
● Detailed recipe pages with images and instructions.
● User profile for customization.
● Conclusion & Future:
● Empowers users, future plans include expanding datasets and enhancing language support.
 Recipe recommendation using ingredient networks(/ui.adsabs.harvard.edu)
Chun-Yuen Teng School of Information University of Michigan Ann Arbor, MI, USA chunyuen@umich.edu
Introduction:
• Online recipes, a staple human tradition, thrive on platforms like Allrecipes.com.
• Our study mines this popular recipe-sharing site to uncover culinary wisdom.
Methodology:
• Ingredient Extraction:
• Two methods used: maximal matching and regular expression.
• Provides a reasonable approximation of ingredients in recipes.
• Ingredient Networks:
• Complement Network:
• Reveals two large communities: savory and sweet.
• Insights into common ingredient pairings.
• Substitute Network:
• Uncovers communities of functionally equivalent ingredients.
• Enhances recipe customization and offers healthier alternatives.
Experiments:
• Predicted recipe ratings with 79.2% accuracy, with 84% prediction power from ingredient networks.

Future Work:
• Refine ingredient extraction and enhance network analysis.
• Explore additional factors influencing recipe ratings.
• Consider user demographics and adapt to changing food trends for improved recommendations.
 Personalized Recipe Recommendation System using Hybrid Approach
Lipi Shah1 , Hetal Gaudani2 , Prem Balani3 (2018)(IJARCCE)
● Introduction:
● Surge in online recipes demands advanced recommendation systems.
● Proposed hybrid model blends content and collaborative filtering for improved accuracy.
● Motivation:
● Current systems struggle with too few ratings and biased popularity
● Our hybrid mix aims to fix this by combining what's in the recipe with what users like
● Proposed Approaches:
● Hybrid Approach-1 (KNN Collaborative Filtering):
○ Integrates content (ingredients, cuisines) with user ratings.
○ Uses K-nearest neighbor algorithm for personalized rating predictions.
● Hybrid Approach-2 (Stochastic Gradient Descent):
○ Utilizes SGD algorithm for model creation using content and user ratings.
○ Minimizes errors for precise rating predictions.
● Evaluation and Results:
● Hybrid models outperform conventional ones.
● Evaluation based on RMSE; Hybrid Approach-2 achieves 0.6365, surpassing conventional models.
● Implications:
● Advances culinary recommendation systems.
● Hybrid models offer precise personalized recipe recommendations, catering to diverse preferences.
 Learning Structural Representations for Recipe Generation and Food Retrieval
Hao Wang, Guosheng Lin, Steven C. H. Hoi, Fellow, IEEE and Chunyan Miao

● Introduction
○ Computers are learning how to help us with cooking, like making recipes and finding food in pictures.
○ Making recipes and finding food in pictures is tricky for computers because recipes are long, and pictures don't show details well.
○ The paper suggests a smart way for computers to understand recipes better. It helps them do a better job in making recipes and finding food in pictures.
● Methodology
● Image-to-Tree:
○ What: Turn food images into hierarchical structures.
○ How: Use a special computer model that learns to arrange image details like a family tree.
● Recipe Creation:
○ What: Make recipes from ingredients and images.
○ How: Teach a computer program (like a smart chef) to write recipes using both pictures and ingredient lists.
● Food Matching:
○ What: Connect food images with their recipes.
○ How: Train a computer to learn what ingredients and steps match with each food picture, so it can find the right recipes.
● Experiments
● Describing long recipes is tough.
● Training with a focus on improving the matching accuracy.
● Using special models for generating recipes and matching images.(ON-LSTM for recipe generation; transformer for cross-modal.)
 Rice Quality Analysis Based On Physical Attributes Using Image Processing
Namita Patel,Assistant Professor,Department Of Information Technology,ASOIT

Introduction:

● Rice is a crucial crop in both human consumption and the food market.
● This review paper explores using image processing techniques and computer vision for rice quality analysis.

Key Components of Image Processing:

● Image processing involves pre-processing, enhancement, display, and information extraction.

● Researchers emphasize that object shape is more informative than texture and color.

Challenges in Rice Quality Analysis:

● One challenge is "rice integrity," specifically, line and point touching of seeds.
● Edge detection methods have advantages and disadvantages, resulting in both real and unreal edges.

Proposed Method:

● The primary aim is to offer an alternative approach for quality control and analysis that reduces effort, cost, and time.
● Image processing plays a crucial role in replacing traditional human sensory panels.

Problem Definition:

● Traditional methods of quality analysis by experienced individuals yield inconsistent results and take time.
● Image processing techniques are proposed to maintain quality specifications consistently.

Rice Quality and Classification:

● Quality analysis has become crucial due to market demands.

● Conventional methods for grain shape and size measurements are slow, costly, and prone to human errors.
 Food Recipe Recommendation Based on Ingredients Detection Using Deep Learning
MD Kishor Morol,American International University,Bangladesh

Introduction:

● Food selection and recipe choice can be a daunting task, especially for beginners.
● Traditional methods for recognizing food ingredients through image processing have limitations.
● Leveraging machine learning and deep learning can offer more accurate solutions.

Methodology:

Data Collection: We compi

led a custom dataset comprising 9856 images from 32 distinct food ingredient classes.
Ingredient Recognition: A Convolutional Neural Network (CNN) model was trained to identify food ingredients from images.
Recipe Recommendation: Machine learning techniques were employed to suggest recipes based on the recognized ingredients.

Results:

● The model achieved an impressive accuracy rate of 94% in recognizing food ingredients.
● The recipe recommendation algorithm provided users with tailored recipe suggestions based on their chosen ingredients.

Conclusion:

● This research presents an innovative solution to the challenges of food ingredient recognition and recipe recommendation.
● Machine learning and deep learning techniques offer a more accurate and efficient approach.
● Our results demonstrate the potential for enhancing the culinary experience for both novice and expert cooks.
 Recipe Recommendation Based on Ingredients using Machine Learning
Ms. Soundarya Desai , Ms. Pooja Patil , Mr. Pratik Shinde , Mr. Azhar Sayyed , Prof. Rohini Bhosale Student, Comp Dept, KJEI‟s Trinity College of
Engineering and Research Pisoli, Pune, India

Introduction:

● Meal planning can be a daily struggle, especially when catering to children's preferences and dietary needs.
● Managing kitchen ingredients efficiently and preventing forgotten purchases is a common challenge.
● Cooking recipes from various sources can be frustrating when key ingredients are missing.

Methodology:

Ingredient Recommendation: We build a recommendation system based on ingredient co-occurrence in recipe databases.
Cooking Ontology: Ingredient categories are stored in a cooking ontology for structured recommendations.
Object Detection with OpenCV: We use OpenCV, a Java-based platform, for feature extraction of ingredients in the kitchen.
Grayscale and Histogram Analysis: We calculate the grayscale of images and generate histograms to identify food items in the kitchen.

Results:

● The ingredient recommendation system assists users in finding suitable ingredients based on what's available.
● Cooking ontology and co-occurrence analysis improve the relevance of ingredient recommendations.
● OpenCV facilitates the detection of ingredients in the kitchen, making it easier to work with available items.

Conclusion:

● This research offers a solution to common cooking dilemmas related to ingredient selection and recipe preparation.
● The recommendation system simplifies meal planning, while object detection enhances kitchen efficiency.
● By combining technology and culinary expertise, this system aims to make cooking a more enjoyable and hassle-free experience.
 Multimodal Recipe Recommendation System Using Deep Learning and Rule-Based Approach
Abdullah Faiz Ur Rahman Khilji,Department of Computer Science and Engineering, National Institute of Technology Silchar, Assam, India

Introduction:

● The digital age provides access to a vast repository of information across interconnected networks.
● However, the unstructured nature of this information necessitates intelligent systems to serve users effectively.
● Recommendation engines today are expected to provide highly satisfactory results by considering various data modalities, particularly text and
images.

Methodology:

Data Collection: A comprehensive dataset is assembled, featuring recipes in textual and image formats.
Deep Learning: The recommendation system employs deep learning techniques to analyze the dataset for optimal recipe recommendations.
Rule-Based Approach: A rule-based approach further enhances the recommendation system's accuracy and relevance in suggesting recipes.
Text and Image Searches: The system offers users the ability to seek information seamlessly through both text and image-based searches.

Results:

● The developed recommendation system simplifies culinary exploration by providing users with relevant recipe suggestions based on their input
ingredients or food images.
● The combination of deep learning and rule-based methodologies ensures accurate and tailored recipe recommendations.
● Users can conveniently explore desired information through text and image-based queries.

Conclusion:

● In a world of vast, unstructured information, recommendation engines play a crucial role in serving user needs.
● The recipe recommendation system presented here is an example of how deep learning and rule-based techniques can be integrated for a more
satisfying user experience.
● By accommodating both text and image searches, this system aims to make culinary exploration a delightful and effortless journey for all users.

n:

● The internet era has unleashed a vast amount of unstructured information, demanding intelligent solutions for data retrieval.
● Recommendation engines now aim to provide highly satisfying services, considering different data types, including text and images.
● This paper introduces a recipe recommendation system that leverages deep learning and rule-based methodologies.
 An effective explainable food recommendation using deep image clustering and community detection
Mehrdad Rostami,Center for Machine Vision and Signal Analysis (CMVS), University of Oulu, Finland
Introduction:

● Visual content in the food domain is pivotal for capturing users' attention and influencing their dietary choices.
● This paper introduces an Explainable Food Recommendation system that combines deep learning and image clustering.
● The recommendation system aims to enhance transparency and interpretability in dietary recommendations.

Methodology:

Deep Learning and Image Clustering: The system employs deep learning techniques to cluster food images and utilize them for dietary recommendations.
Novel Similarity Score: A novel similarity score, based on user community preferences for food categories, is introduced to refine the recommendation process.
Rule-Based Explainability: The recommendation system includes a rule-based explainability component to make the recommendations more transparent and
interpretable.
Experiments and Ablation Study: Experiments on a collected dataset demonstrate the substantial improvement in recommendation quality, and an ablation
study confirms the technical soundness of the system components.

Results:

● The proposed recommendation system significantly improves recommendation quality, as evidenced by increases in precision, recall, F1 score, and NDCG.
● The rule-based explainability feature enhances transparency and interpretability in the recommendations.

Conclusion:

● This research highlights the importance of visual content in dietary communication.

● The Explainable Food Recommendation system showcases the power of combining deep learning, novel similarity scoring, and rule-based explainability.
● The system offers a promising approach to making food recommendations that are not only more accurate but also more transparent and interpretable,
ultimately influencing users' dietary choices positively.
 Conclusion

In conclusion, our recipe app isn't just about technology—it's a culinary revolution. With
cutting-edge machine learning, we personalize your kitchen experience, adapting to your tastes.
Beyond recipes, we address nutritional concerns, aiming to empower informed dietary choices by
accurately quantifying calorie content.
Embark on this transformative culinary journey with us. It's not just about following recipes; it's
discovering a world of tailored dishes crafted uniquely for you. Join us in reshaping cooking, where
technology meets flavor, and every meal becomes a personalized adventure. Welcome to the
future of culinary exploration.
Thank You!!